entitygraph

Documentation

Overview

Package entitygraph implements the entity and source graph runtime. Extraction primitives and dedup logic; Entity/Mention types live in internal/store to avoid a circular import (entitygraph → parser → store).

Index

• Constants
• func AllowedTypes(packs []domainpacks.Pack) map[string]bool
• func Dedup(result *ExtractResult)
• func IndexFile(st EntityWriter, relPath string, res *parser.ParseResult) error
• func NormalizeName(name string) string
• type EntityWriter
• type ExtractResult
  • func FromParseResult(res *parser.ParseResult, allowedTypes map[string]bool) ExtractResult

Constants

const (
	MaxAliases    = 200
	MaxContextLen = 500
	MaxEntities   = 500
)

Functions

func AllowedTypes

func AllowedTypes(packs []domainpacks.Pack) map[string]bool

AllowedTypes returns the set of allowed entity_type values drawn from packs whose Domain == "entity". Vocabulary values are taken from the Aliases of any Field with Key == "entity_type". An empty map means no vocabulary constraint — all entity types declared in frontmatter are accepted.

func Dedup

func Dedup(result *ExtractResult)

Dedup merges duplicate entities in result by (entity_type, canonical_name_normalized). The UUID PK of the first-seen entry is kept so FK references remain stable. Aliases from all duplicate entries are merged (deduplicated, capped at MaxAliases). Mention.EntityID values are rewritten to point to the surviving entity.

func IndexFile

func IndexFile(st EntityWriter, relPath string, res *parser.ParseResult) error

IndexFile extracts entity data from one parsed document and persists it. Called per changed file inside the indexStore loop after metadata upserts.

func NormalizeName

func NormalizeName(name string) string

NormalizeName returns a lowercase-trimmed canonical form of name. Internal whitespace runs are collapsed to a single space.

Types

type EntityWriter

type EntityWriter interface {
	InsertEntities([]store.Entity) error
	InsertEntityMentions([]store.Mention) error
}

EntityWriter is the narrow persistence surface IndexFile needs: just the two entity-graph writes. Depending on this interface (rather than the concrete *store.Store) keeps entitygraph decoupled from the rest of the store API and makes IndexFile trivially mockable. *store.Store satisfies it via the thin forwarders in internal/store/forwarders.go.

type ExtractResult

type ExtractResult struct {
	Entities []store.Entity
	Mentions []store.Mention
}

ExtractResult is the output of extraction from a single ParseResult.

func FromParseResult

func FromParseResult(res *parser.ParseResult, allowedTypes map[string]bool) ExtractResult

FromParseResult extracts entity candidates from a single parsed document. Sources: explicit frontmatter "entities:" list and wikilinks whose targets match entries in allowedTypes (or all wikilinks when allowedTypes is empty). No LLM, network, or file I/O.